Quality Evaluation Algorithm for Chest Compressions Based on OpenPose Model

Abstract

Aiming at the problems of the low evaluation efficiency of the existing traditional cardiopulmonary resuscitation (CPR) training mode and the considerable development of machine vision technology, a quality evaluation algorithm for chest compressions (CCs) based on the OpenPose human pose estimation (HPE) model is proposed. Firstly, five evaluation criteria are proposed based on major international CPR guidelines along with our experimental study on elbow straightness. Then, the OpenPose network is applied to obtain the coordinates of the key points of the human skeleton. The algorithm subsequently calculates the geometric angles and displacement of the selected joint key points using the detected coordinates. Finally, it determines whether the compression posture is standard, and it calculates the depth, frequency, position and chest rebound, which are the critical evaluation metrics of CCs. Experimental results show that the average accuracy of network behavior detection reaches 94.85%, and detection speed reaches 25 fps.